Uniform Forward-Modeling Analysis of Ultracool Dwarfs. I. Methodology and Benchmarking

TL;DR

This paper introduces a Bayesian forward-modeling framework using the Starfish tool and cloudless Sonora-Bobcat models to analyze low-resolution near-infrared spectra of T dwarfs, improving parameter estimation accuracy.

Contribution

It develops a novel Bayesian forward-modeling methodology that accounts for uncertainties and systematics, providing more realistic atmospheric and physical parameters for ultracool dwarfs.

Findings

Accurate effective temperatures and radii for some benchmarks.

Underestimated surface gravities and metallicities due to model systematics.

Masses derived from spectral fits are significantly underestimated.

Abstract

We present a forward-modeling framework using the Bayesian inference tool Starfish and cloudless Sonora-Bobcat model atmospheres to analyze low-resolution ($R\approx80-250$) near-infrared ($1.0-2.5$ $\mu$m) spectra of T dwarfs. Our approach infers effective temperatures, surface gravities, metallicities, radii, and masses, and by accounting for uncertainties from model interpolation and correlated residuals due to instrumental effects and modeling systematics, produces more realistic parameter posteriors than traditional ($\chi^2$-based) spectral-fitting analyses. We validate our framework by fitting the model atmospheres themselves and finding negligible offsets between derived and input parameters. We apply our methodology to three well-known benchmark late-T dwarfs, HD 3651B, GJ 570D, and Ross 458C, using both solar and non-solar metallicity atmospheric models. We also derive these benchmarks' physical properties using their bolometric luminosities, their primary stars' ages and metallicities, and Sonora-Bobcat evolutionary models. Assuming the evolutionary-based parameters are more robust, we find our atmospheric-based, forward-modeling analysis produces two outcomes. For HD 3615B and GJ 570D, spectral fits provide accurate $T_{\rm eff}$ and $R$ but underestimated $\log{g}$ (by $\approx1.2$ dex) and $Z$ (by $\approx0.35$ dex), likely due to the systematics from modeling the potassium line profiles. For Ross 458C, spectral fits provide accurate $\log{g}$ and $Z$ but overestimated $T_{\rm eff}$ (by $\approx120$ K) and underestimated $R$ (by $\approx1.6\times$), likely because our model atmospheres lack clouds, reduced vertical temperature gradients, or disequilibrium processes. Finally, the spectroscopically inferred masses of these benchmarks are all considerably underestimated.